4.5. Confidence and Validation
Although climate models should aid understanding in the processes which control and perturb the climate, the confidence placed in such models should always be questioned. Critically, it should be remembered that all climate models represent a simplification of the climate system, a system which indeed may ultimately prove to be too complex to model.
Given that many of the processes which are modelled occur over time scales so long that it is impossible to test model results against real-world observations, it is also arguable that climate modelling is, in some respects. philosophically suspect. Model performance can be tested through the simulations of shorter time scale processes but short-term performance may not necessarily reflect long-range accuracy.
Climate models must therefore be used with care and their results interpreted with due caution. Margins of uncertainty must be attached to any model projection. Uncertainty margins can be derived by the comparison of the results of different model experiments or through sensitivity studies, in which key assumptions are altered to determine the role they play in influencing the final climatic response.
Validation of climate models (testing against real-world data) provides the only objective test of model performance. As far as GCMs are concerned, validation exercises have revealed a number of deficiencies in their simulations of present-day conditions (e.g. Gates et al., 1990):
a) modelled stratospheric temperatures tend to be lower than equivalent instrumental observations;
b) modelled mid-latitude westerlies tend to be too strong; easterlies are too weak;
c) modelled sub-polar low pressure systems in winter tend to be too deep and displaced east, and;
d) day-to-day variability is lower than in the real world.
Finally, it has been observed that some models suffer from climate drift. The background climate shifts as the simulation proceeds, despite the absence of any climate forcing.